The issue with architectural decisions is that they effect the whole system and/or you often need to make them early in the development process. It means a lot effort if you change that decision a couple of months later. From an economic standpoint architectural decisions are often irrevocable. Good architecture is one that allows an architect to make late decisions without superior effect on efforts and costs. Let's put that on record.
Law 1: Good architecture is one that enables architects to have a minimum of irrevocable decisions.
To minimize the set of irrevocable decisions the system needs to be responsive to change. There is a major lesson I have learned about software development projects: Nothing is permanent except change. The client changes his opinion about requirements. The stakeholders change their viewpoint of what's important. People join and leave the project team. The fact that change alone is unchanging leads me to the second rule of good architecture, that is:
Law 2: To make decisions revocable you need to design for flexibility.
This is the most provocative statement and I am having controversial discussions here. The reason is that flexibility introduces the need for abstraction. Abstraction uses a strategy of simplification, wherein formerly concrete details are left ambiguous, vague, or undefined (from Wikipedia). This simplification process isn't always simple to do and to follow for others in particular. "Making something easy to change makes the overall system a little more complex, and making everything easy to change makes the entire system very complex. Complexity is what makes software hard to change." from M. Fowler) This is one core problem of building good software architecture: Developing software that is easy to change but at the same time understandable. There are several concepts that try to tackle this paradox problem: design patterns and object oriented design principles. Polymorphism, loose coupling and high cohesion are flexibility enablers to me.
Law 3: To make use of flexibility one needs to refactor mercilessly.
Flexibility is not an end in itself. You need to actively make use of flexible design. If something is changing and it makes a previous design or architectural decision obsolete you need to go into the code and change the software. Otherwise the effort of building flexible software is useless and technical debt may cause late delays and a maintenance nightmare. The fact that you take rigorous action on your code base requires continuous feedback about the qualities of your software. To be able to refactor it is therefore essential that the code base is covered by a sufficient amount of automated tests. In an ideal scenario everything is integrated into a continuous integration environment to receive permanent feedback about the health of your code base.